ibm-aimc Phi-3 and Llama-3.2 Weights #834

Missing Pre-trained weights for phi-3 and llama-3.2

Hi, thanks for the amazing work here. Codes are very straightforward to understand. I wonder if you are still planning or about to release these two weights by chance?

hf download ibm-aimc/analog-Phi-3-mini-4k-instruct
--local-dir /app/code/naihao_summer/models/ibm-aimc/analog-Phi-3-mini-4k-instruct
hf download ibm-aimc/analog-Llama-3.2-1B-Instruct
--local-dir /app/code/naihao_summer/models/ibm-aimc/analog-Llama-3.2-1B-Instruct

I tried to use the same set to simulate your clipping range using the exact same code with boolq set and evaluated on the other zero-shot sets, but the results were terrible (20-30% lower for all zero-shot dataset evaluation and MedQA). Wonder if it's possible that you can give some information on the datasets that you used to get the weights above.

Appreciated